The Klein 1989 pre-mortem method asks teams to imagine the project has failed and reason backward to causes. It works, sometimes. The version most teams run is decorative: failure modes get named in a meeting, written into a doc, and then ignored when the project ships. This skill ships a pre-mortem with the forcing functions that make the method bind on actual behavior. What's in the file: - Five-phase procedure (premortem framing, free generation of failure modes, behavioral attractor test, probability/impact assessment, pre-commitment design) with explicit steps each phase must complete before advancing. - Behavioral attractor test (Phase 3): after generating failure modes, remove the highest-probability one and re-derive. If the same structural failure re-emerges with different surface details, the team has a behavioral attractor not a discrete risk. The skill writes a literal "FORBIDDEN FROM RE-DERIVATION: " line into context as an in-context suppression anchor for the LLM running the analysis. - Ordinal probability ranking (Phase 4): HIGH/MEDIUM/LOW × Existential/Severe/Recoverable/Minor, anchored against named reference classes. No fake-precision point estimates. - Pre-commitment enforcement layer (Phase 5): every pre-commitment device requires three fields explicitly written down. NAMED OWNER (the person who personally answers if the device is breached), EXTERNAL WITNESS (someone outside the team who is told what the commitment is), and RENEGOTIATION CLAUSE (the conditions under which the commitment can be dropped without it being a quiet failure). Without these three fields, the device is decorative. - Worked example demonstrating all four mechanics on a realistic B2B SaaS launch scenario. - META-REFLECTION self-check inlined into OUTPUT DISCIPLINE step 0 as a hard gate. - Anti-patterns section catching common pre-mortem failures (planning-fallacy laundering, narrative compliance in Phase 3, pre-commitments without enforcement). Built for: founders running pre-launch reviews, project managers running risk sessions, tech leads running architecture decisions where the cost of being wrong is high, and LLM-augmented planning tools that need a procedure that fires per-step rather than once at the top. Cites: Klein 2007 HBR, Kahneman planning fallacy, Tetlock calibration, Hubbard probability estimation, Annie Duke pre-commitment.